reddit · SaaS · 09-17 14:50 65 分 ← 返回列表

How do you know which customer is burning your OpenAI budget?

Once an AI bill stops being a rounding error, the first question is "which customers is that", and the provider dashboard has no concept of your tenants. Instrumenting every call site with logging is the obvious fix and nobody wants to do it. What I built is a proxy: swap the base URL, keep your own provider key, add a customer id header and a feature name header on each request. The proxy reads the usage block, prices it against a rate table, and the report groups by customer, feature or m

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：多租户 SaaS 的 OpenAI/AI API 账单增长后，无法按客户、租户或功能归因成本，导致定价和成本控制困难。

目标用户：使用 OpenAI/AI API 并按席位或用量计费的多租户 B2B SaaS 团队。

证据：Once an AI bill stops being a rounding error, the first question is "which customers is that", and the provider dashboard has no concept of your tenants.

变现：按监控的 API 调用量或月订阅收费，提供分层定价；企业版可支持自托管、自定义费率表、SSO 和告警。

MVP：一个反向代理：替换 base_url，保留自己的 provider key，在请求中注入 customer_id 和 feature 头，解析 OpenAI usage，按自定义费率表计算，生成按客户、功能、模型分组的成本报表并支持导出。

竞品：Helicone、LiteLLM、Portkey、OpenMeter、Langfuse、OpenAI usage dashboard

摘要：发帖人构建了一个 OpenAI 代理，用于按客户、功能、模型归因 AI 成本，解决多租户 SaaS 账单翻倍后无法定位到具体租户的问题。目前没有明确付费意愿或收入数据，但痛点真实，独立开发者可快速做 MVP，竞争来自 LLM 可观测性与网关工具。